Deep Neural Network-based Detection and Partial Response Equalization for Multilayer Magnetic Recording

Aboutaleb, Ahmed; Sayyafan, Amirhossein; Belzer, Benjamin J.; Sivakumar, Krishnamoorthy; Greaves, S.J.; Chan, Kheong Sann; Wood, Roger

doi:10.1109/tmrc49521.2020.9366719

Cited by 2 publications

(3 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Despite the improvements in performance reported in [5]- [8], the mentioned NN-based equalizers are much more complex than the linear equalizer baseline. Indeed, the high complexity of NN-based methods precludes practical implementation.…”

Section: Introductionmentioning

confidence: 99%

“…Compared with conventional equalizers and detection systems, neural networks (NNs) have been shown to compensate better for the non-linear impairments in high density magnetic recording channels with significant improvement in the overall system performance for 1DMR in [2], TDMR in [3], [5]- [7], and multilayer magnetic recording (MLMR) in [8].…”

Section: Introductionmentioning

confidence: 99%

“…In addition, for application in TDMR, in [5], Shen et al propose a detection system consisting of a 2D-LMMSE followed by a CNN detector. In [8], Aboutaleb et al propose CNN-based systems for detection and equalization in MLMR channels with severe non-linear distortions. In [3], Shen and Nangare propose an NN equalizer followed by a VA, where the NN's parameters adapt to minimize the cross-entropy (CE).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Reduced Complexity Neural Network Equalizers for Two-dimensional Magnetic Recording

Aboutaleb¹,

Nangare²

2021

Preprint

Self Cite

View full text Add to dashboard Cite

Recent studies show promising performance gains achieved by non-linear equalization using neural networks (NNs) over traditional linear equalization in two-dimensional magnetic recording (TDMR) channels. But the examined neural network architectures entail much higher implementation complexities compared with the linear equalizer, which precludes practical implementation. For example, among the low complexity reported architectures, the multilayer perceptron (MLP) requires about 6.6 times increase in complexity over the linear equalizer. This paper investigates candidate reduced complexity neural network architectures for equalization over TDMR. We test the performance on readback signals measured over an actual hard disk drive with TDMR technology. Four variants of a reduced complexity MLP (RC-MLP) architecture are proposed. A proposed variant achieves the best balance between performance and complexity. This architecture consists of finite-impulse response filters, a nonlinear activation, and a hidden delay line. The complexity of the architecture is only 1.59 times the linear equalizer's complexity, while achieving most of the performance gains of the MLP.

show abstract

Section: Introductionmentioning

confidence: 99%